| 摘要 |
The invention relates, in the field of subband decomposition, to the design of filter banks adapted to the input signal statistics. In most cases, two channel filter banks are iteratively applied over several levels of decomposition, the signals in the resulting subbands representing decimated and filtered versions of the input signal. According to the invention, it is proposed a perfect reconstruction critically decimated polyphase filter bank with a ladder structure, which adapts to the nonstationarities in the input signal. In the simplest embodiment, four steps are provided in the implementation of the filtering method: a splitting step (21), provided for subdividing the input signal c0(n) into two disjoint subsets c0(2n) and c0(2n+1) of samples (odd and even ones), a predicting step (22), provided for predicting on the basis of the even subset the odd one (d1(n)=c0(2n+1)-P1[c0(2n)], an updating step (23), provided for generating on the basis of said predicted odd subset the even one (c1(n)=c0(2n)+U1[d1(n)], and an iterative cross-optimization step associating said updating step (23) of the current decomposition level and the predicting step (32) of the following one. More generally, several successive similar scales may be provided, a minimization of the variance of the obtained coefficients being used as an optimization criterion at each scale.
|
